CoEdIT Base — ONNX
ONNX export of jbochi/coedit-base (250M parameters) with encoder-decoder architecture and KV cache support.
CoEdIT is a T5-based model fine-tuned on the grammarly/coedit dataset for text editing tasks including grammar correction, simplification, coherence, and paraphrasing. This base variant is fine-tuned from google/flan-t5-base.
Converted for use with inference4j, an inference-only AI library for Java.
Original Source
- Repository: jbochi/coedit-base
- License: Apache 2.0
Usage with inference4j
try (var corrector = CoeditGrammarCorrector.coeditBase().build()) {
System.out.println(corrector.correct("She don't likes swimming."));
// She doesn't like swimming.
}
Model Details
| Property | Value |
|---|---|
| Architecture | T5 encoder-decoder (250M parameters) |
| Base model | google/flan-t5-base |
| Training data | grammarly/coedit |
| Task | Grammar correction, text editing |
| Tokenizer | SentencePiece (32,128 tokens) |
| Original framework | PyTorch (transformers) |
| Export method | Hugging Face Optimum (encoder-decoder with KV cache) |
License
This model is licensed under the Apache License 2.0. Original model by jbochi, trained on the Grammarly CoEdIT dataset.